Context-Aware Research Brief Generator

A research assistant that generates structured, evidence-linked briefs with follow-up support using LangGraph and LangChain.
Features a deployed API, CLI, and optional Streamlit UI.

• Deployment: Live App
• Repository: GitHub Repo

Problem Statement and Objective

This project creates research briefs from user topics with web evidence, supporting follow-up queries with context.
It leverages:

• LangGraph for orchestration
• LangChain for abstraction
• Pydantic for schema validation

System Architecture

The system is designed as a graph-based workflow with modular nodes.

Mermaid Graph Representation

graph TD A[User Input: API/CLI] --> B[Context SummarizationNode] A --> C[Planning Node] B --> C C --> D[Search Node(Tavily API)] D --> E[Content Fetching Node(Requests, BeautifulSoup)] E --> F[Per-Source Summarization Node(Gemini LLM)] F --> G[Synthesis Node(Gemini LLM)] G --> H[Post-Processing Node] H --> I[Output: FinalBrief JSON] I --> J[Streamlit UI(Local)] I --> K[SQLite History] L[Checkpointing(MemorySaver)] --> B L --> C L --> D L --> E L --> F L --> G L --> H

Node Descriptions

• Context Summarization → Condenses input for planning.
• Planning → Breaks down tasks into structured steps.
• Search (Tavily API) → Web retrieval with relevance filtering.
• Content Fetching → Extracts raw content via requests + BeautifulSoup.
• Per-Source Summarization → Summarizes each source using Gemini LLM.
• Synthesis → Combines summaries into a unified research brief.
• Post-Processing → Cleans, validates, and formats output.
• FinalBrief JSON → Structured output stored in SQLite and viewable via Streamlit.
• Checkpointing (MemorySaver) → Ensures fault tolerance and recoverability.

Model and Tool Selection

• Gemini (gemini-1.5-flash) → Summarization & synthesis (fast + cost-efficient).
• Tavily → Efficient web search with relevance ranking.
• BeautifulSoup → Reliable content parsing.